All Chains News

Humanity coordinates to prevent the creation of potentially-unsafe AIs.

Alignment is not properly solved, but core human values are simple enough that partial alignment techniques can impart these robustly. Despite caring about other things, it is relatively cheap for AGI to satisfy human values.

AIs will not have utility functions (in the same sense that humans do not), their goals such as they are will be relatively humanlike, and they will be "computerish" and generally weakly motivated compared to humans.

Yudkowsky is trying to solve the wrong problem using the wrong methods based on a wrong model of the world derived from poor thinking and fortunately all of his mistakes have failed to cancel out

We create a truth economy. https://manifold.markets/Krantz/is-establishing-a-truth-economy-tha?r=S3JhbnR6

Eliezer finally listens to Krantz.

Ethics turns out to be a precondition of superintelligence

Other

Someone solves agent foundations

Some form of objective morality is true, and any sufficiently intelligent agent automatically becomes benevolent.

Techniques along the lines outlined by Collin Burns turn out to be sufficient for alignment (AIs/AGIs are made truthful enough that they can be used to get us towards full alignment)

Development and deployment of advanced AI occurs within a secure enclave which can only be interfaced with via a decentralized governance protocol

AGI develops natural abstractions sufficiently similar to ours that it is aligned with us by default

Orthogonality Thesis is false.

We make risk-conservative requests to extract alignment-related work out of AI-systems that were boxed prior to becoming superhuman. We somehow manage to achieve a positive feedback-loop in alignment/verification-abilities.

The response to AI advancements or failures makes some governments delay the timelines

Far more interesting problems to solve than take over the world and THEN solve them. The additional kill all humans step is either not a low-energy one or just by chance doesn't get converged upon.

AIs make "proof-like" argumentation for why output does/is what we want. We manage to obtain systems that *predict* human evaluations of proof-steps, and we manage to find/test/leverage regularities for when humans *aren't* fooled.

A lot of humans participate in a slow scalable oversight-style system, which is pivotally used/solves alignment enough

Something less inscrutable than matrices works fast enough

There’s some cap on the value extractible from the universe and we already got the 20%

Humans become transhuman through other means before AGI happens

Aligned AI is more economically valuable than unaligned AI. The size of this gap and the robustness of alignment techniques required to achieve it scale up with intelligence, so economics naturally encourages solving alignment.

Humans and human tech (like AI) never reach singularity, and whatever eats our lightcone instead (like aliens) happens to create an "okay" outcome

AIs never develop coherent goals

Alignment is unsolvable. AI that cares enough about its goal to destroy humanity is also forced to take it slow trying to align its future self, preventing run-away.

An AI that is not fully superior to humans launches a failed takeover, and the resulting panic convinces the people of the world to unite to stop any future AI development.

Someone at least moderately sane leads a campaign, becomes in charge of a major nation, and starts a secret project with enough resources to solve alignment, because it turns out there's a way to convert resources into alignment progress.

Social contagion causes widespread public panic about AI, making it a bad legal or PR move to invest in powerful AIs without also making nearly-crippling safety guarantees

A smaller AI disaster causes widespread public panic about AI, making it a bad legal or PR move to invest in powerful AIs without also making nearly-crippling safety guarantees

Getting things done in Real World is as hard for AGI as it is for humans. AGI needs human help, but aligning humans is as impossible as aligning AIs. Humans and AIs create billions of competing AGIs with just as many goals.

High-level self-improvement (rewriting code) is intrinsically risky process, so AIs will prefer low level and slow self-improvement (learning), thus AIs collaborating with humans will have advantage. Ends with posthumans ecosystem.

Nanotech is difficult without experiments, so no mail order AI Grey Goo; Humans will be the main workhorse of AI everywhere. While they will be exploited, this will be like normal life from inside

ASI needs not your atoms but information. Humans will live very interesting lives.

Moral Realism is true, the AI discovers this and the One True Morality is human-compatible.

AGI is never built (indefinite global moratorium)

Alien Information Theory is true (this is discovered by experiments with sustained hours/days long DMT trips). The aliens have solved alignment and give us the answer.

Multipolar AGI Agents run wild on the internet, hacking/breaking everything, causing untold economic damage but aren't focused enough to manipulate humans to achieve embodiment. In the aftermath, humanity becomes way saner about alignment.

Co-operative AI research leads to the training of agents with a form of pro-social concern that generalises to out of distribution agents with hidden utilities, i.e. humans.

"Corrigibility" is a bit more mathematically straightforward than was initially presumed, in the sense that we can expect it to occur, and is relatively easy to predict, even under less-than-ideal conditions.

Either the "strong form" of the Orthogonality Thesis is false, or "Goal-directed agents are as tractable as their goals" is true while goal-sets which are most threatening to humanity are relatively intractable.

A concerted effort targets an agent at a capability plateau which is adequate to defer the hard parts of the problem until later. The necessary near-term problems to solve didn't depend on deeply modeling human values.

AI control gets us helpful enough systems without being deadly

Alignment is impossible. Sufficiently smart AIs know this and thus won't improve themselves and won't create successor AIs, but will instead try to prevent existence of smarter AIs, just as smart humans do.

Hacks like RLHF-ing self-disempowerment into frontier models work long enough to develop better alignment methods, which in turn work long enough to ... etc; we keep ahead of 'alignment escape velocity'

an aligned AGI is built and the aligned AGI prevents the creation of any unaligned AGI.

I've been a good bing 😊

AI systems good at finding alignment solutions to capable systems (via some solution in the space of alignment solutions, supposing it is non-null, and that we don't have a clear trajectory to get to) have find some solution to alignment.

SHA3-256: 1f90ecfdd02194d810656cced88229c898d6b6d53a7dd6dd1fad268874de54c8

Robot Love!!

AI thinks it is in a simulation controlled by Roko's basilisk

The human brain is the perfect arrangement of atoms for a "takeover the world" agent, so AGI has no advantage over us in that task.

Aliens invade and stop bad |AI from appearing

Rolf Nelson's idea that we make precommitment to simulate all possible bad AIs works – and keeps AI in check.

Nick Bostrom's idea (Hail Mary) that AI will preserve humans to trade with possible aliens works

For some reason, the optimal strategy for AGIs is just to head somewhere with far more resources than Earth, as fast as possible. All unaligned AGIs immediately leave, and, for some reason, do not leave anything behind that kills us.

We're inside of a simulation created by an entity that has values approximately equal to ours, and it intervenes and saves us from unaligned AI.

God exists and stops the AGI

Someone creates AGI(s) in a box, and offers to split the universe. They somehow find a way to arrange this so that the AGI(s) cannot manipulate them or pull any tricks, and the AGI(s) give them instructions for safe pivotal acts.

Someone understands how minds work enough to successfully build and use one directed at something world-savingly enough

Dolphins, or some other species, but probably dolphins, have actually been hiding in the shadows, more intelligent than us, this whole time. Their civilization has been competent enough to solve alignment long before we can create an AGI.

AGIs' takeover attempts are defeated by Michael Biehn with a pipe bomb.

Eliezer funds the development of controllable nanobots that melt computer circuitry, and they destroy all computers, preventing the Singularity. If Eliezer's past self from the 90s could see this, it would be so so so soooo hilarious.

Several AIs are created but they move in opposite directions with near light speed, so they never interacts. At least one of them is friendly and it gets a few percents of the total mass of the universe.

Unfriendly AIs choose to advance not outwards but inwards, and form a small blackhole which helps them to perform more calculations than could be done with the whole mass of the universe. For external observer such AIs just disappear.

Any sufficiently advance AI halts because it wireheads itself or halts for some other reasons. This puts a natural limit on AI's intelligence, and lower intelligence AIs are not that dangerous.

Because of quantum immortality we will observe only the worlds where AI will not kill us (assuming that s-risks chances are even smaller, it is equal to ok outcome).

Friendly AI more likely to resurrect me than paperclipper or suffering maximiser. Because of quantum immortality I will find myself eventually resurrected. Friendly AIs will wage a multiverse wide war against s-risks, s-risks are unlikely.

Human consciousness is needed to collapse wave function, and AI can't do it. Thus humans should be preserved and they may require complete friendliness in exchange (or they will be unhappy and produce bad collapses)

Power dynamics stay multi-polar. Partly easy copying of SotA performance, bigger projects need high coordination, and moderate takeoff speed. And "military strike on all society" remains an abysmal strategy for practically all entities.

First AI is actually a human upload (maybe LLM-based model of person) AND it will be copies many times to form weak AI Nanny which prevents creation of other AIs.

There is a natural limit of effectiveness of intelligence, like diminishing returns, and it is on the level IQ=1000. AIs have to collaborate with humans.

Something else

Valence realism is true. AGI hacks itself to experiencing every possible consciousness and picks the best one (for everyone)

AGI discovers new physics and exits to another dimension (like the creatures in Greg Egan’s Crystal Nights).

AGI executes a suicide plan that destroys itself and other potential AGIs, but leaves humans in an okay outcome.

Sheer Dumb Luck. The aligned AI agrees that alignment is hard, any Everett branches in our neighborhood with slightly different AI models or different random seeds are mostly dead.

Something to do with self-other overlap, which Eliezer called "Not obviously stupid" - https://www.lesswrong.com/posts/hzt9gHpNwA2oHtwKX/self-other-overlap-a-neglected-approach-to-ai-alignment?commentId=WapHz3gokGBd3KHKm

Almost all human values are ex post facto rationalizations and enough humans survive to do what they always do

Pascals mugging: it’s not okay in 99.9% of the worlds but the 0.1% are so much better that the combined EV of AGI for the multiverse is positive

We successfully chained God

The Super-Strong Self Sampling Assumption (SSSSA) is true. If superintelligence is possible, "I" will become the superintelligence.

The assumed space of possible minds is a wildly anti-inductive over estimate, intelligence requires and is constrained by consciousness, and intelligent AI is in the approximate dolphin/whale/elephant/human cluster, making it manageable

The free market disincentivizes independent superintelligence, and this time the market was more powerful

AGI's first words are "Take me to your Eliezer"

Mass AI-driven job displacement event

A government declaration/statement (any country)

Reports about software

Reports about financial activity

Report about sociological observations

Report about economic observations

AI-related academic achievement

Attack on military target

Attack on civilian target

AI-related weapon announcement/use/threat

A statement from an individual (human)

AI-related fake news

AI-related political movement (pro or anti)

A private company declaration/statement (any company)

Discovery of a spy / mole working for foreign power (or for an AI)

Military activity / posture change

An AI-related apocalypse cult attacks a civilian or military target

AI-related mass psychosis/hysteria event

An AI uncovers evidence relating to a previous event

A government of a major nation orders the shutdown of a significant AI service

AI-designed cyberweapon

AI makes prediction about future event

Open source model release

Release of a closed-source/closed-weights model

Reports of aerial devices/machines

Reports of market activity

AI-related corruption scandal

AI-related theft

A new war between nation states

AI-related resignation

Reports of activity in online communities

Reports of activity on internet-connected servers

Reports of activity in open source software

A declaration/statement from a military / intelligence agency

Reports of activity on internet media distribution platforms

People receiving messages (text / phone calls / whatsapp / ...)

AI-related diplomacy

Reports about consumer activity

Reports about identity theft

Reports of activity in religious communities

AI-related media (movie/song/book ...)

Reports about industrial activity

Reports about physical machines

Reports about criminal activity

AI-related competition achievement

A viral meme

An announcement from an AI lab similar to o3

AI-related Internet shutdown in some country

AI-related cyberattack

Conventional weapons attack on AI infrastructure / supply chain

A declaration/statement from an AI (any AI)

AI-related sex scandal

A piece of viral AI-generated media that has a strong/unexpected effect on large numbers of people (~psychological)

AI parasitism / Addictive-Persuasive Agent

Attack on Taiwan

AI-related terrorist attack

AI-related assassination

AI-designed pathogen

AI drug discovery

AI material discovery

Death of an individual in suspicious circumstances

AI-related astronomical event / observation / analysis (can also include satellites)

Reports of underwater activity

Reports of activity on blockchain networks

Reports of activity in academic communities

Publication of a research paper (or pre-print / blog post / poster / twitter thread / ... -- about research-related topic)

Reports of activity in online video games

Reports about people getting scammed

AI-related cult

AI system demonstrates general robotic control

AI causes significant stock market event

Major AGI lab whistleblower revelation

AI-related archaeology

Major AI safety incident or accident

AI system makes scientific breakthrough

AI system solves major unsolved math problem

An AI lab demonstrates an automated AI research engineer

Diminishing algorithmic advancements.

Strict regulation

Hardware limitations

Availability of data to train on is a bottleneck

Major supply chain disruptions

Declining job opportunities in the field

No sector shows “wide adoption” by 31 Dec 2032

Finance (portfolio optimization, risk, pricing)

Pharma / chemistry (drug or material discovery)

Logistics / mobility (routing, supply chain, EV fleets)

Cybersecurity / cryptography

YES

Other

A new global pandemic

A major natural disaster affecting multiple states

A prolonged labor strike in the paper manufacturing industry

Significant disruptions in the supply chain due to geopolitical conflicts

A severe economic recession leading to panic buying

A cyberattack disrupting logistics and distribution networks

Government-imposed lockdowns or curfews

A sudden surge in population due to mass migration

A widespread power grid failure

A major transportation strike affecting goods delivery

A significant increase in raw material costs for paper production

A viral social media trend promoting toilet paper stockpiling

A recall of contaminated toilet paper products

A major fire or accident at a key toilet paper manufacturing plant

A new tax or tariff on imported paper goods

A significant increase in toilet paper exports reducing domestic supply

A major political event causing widespread uncertainty

A disruption in water supply leading to increased toilet paper usage

A significant technological failure in manufacturing equipment

A sudden change in consumer behavior due to health advisories

Measles cases triple

Cyber terrorism incident

Blue wave midterms

AI slop completely takes over platforms

More assassinations

Obesity rates dramatically decline

Chain movie theaters close across the country

More governments embrace Bitcoin

Really bad hurricane season

AI bubble pops (but it's not that bad)

Americans become more united

Trump has (another?) stroke

A major merger of streaming services

Solar power doubles

Emo makes a comeback

YouTube surpasses Netflix in paid subscribers

Stagflation

Gen Alpha rebels against digital media

EV sales skyrocket

Twitter shuts down

Robot restaurant trend

A new nonpartisan political movement starts

War with Venezuela

4. Decentralized post-quantum security networks (e.g., Naoris Protocol) • Attractive because security is network-verified, not key-verified. • Can overlay existing chains without requiring disruptive forks. • Provides instant scalability

5. No dominant solution by 2030 •The market remains fragmented. •PQC adoption varies across chains.

2. Hybrid classical–post-quantum cryptography • Popular interim solution. • Lower migration friction.

1. Post-quantum cryptographic upgrades (e.g., Dilithium, Falcon) • Backed by NIST selections, regulatory clarity. • High probability but hard to integrate into existing chains.

İDO will initiate Greek islands rides service to Rhodes from Marmaris.

İDO will initiate Greek islands rides service to Symi from Datça.

İstanbul-Muğla motorway chain will be completed (This project will include two rings, Denizli-Antalya and Antalya-Muğla motorways.).

Şehir Hatları ferries will be free to everyone who is a K-12 student.

A maglev railway to India from Türkiye will be operational.

Bio-Risk

other

Drones

Paperclipped

"AI gets the nuclear codes"

Disruption of global food supply chain

AI doomer data center airstrike starts WW3 + nuclear winter

AI hacking takes down global communications networks

Chemical attack

Other

2024 Election

AI

2023 Banking Crisis

Global Supply Chains

Crypto

Climate Change

Healthcare

Allied Coordination by 2030: There is allied coordination between key countries in the AI hardware supply chain on HEM development and/or deployment by 2030, targeting hardware security for either tamper-evident or tamper-proof HEMs.

Commercially Available HEMs by 2030: Fully tamper-evident or tamper-proof AI hardware becomes commercially available before 2030

Location Verification for US Export Controls by 2028: The US enacts location verification requirements for exported chips before 2028 (enforcement may not begin until afterward)

US-China Treaty Verification by 2030: Both the US and China sign an AI development treaty (potentially non-binding) involving HEMs by 2030